A Progressive Learning Approach to Chinese SRL Using Heterogeneous Data

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 2069–2077 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1189 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 2069–2077 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1189 A Progressive Learning Approach to Chinese SRL Using Heterogeneous Data Qiaolin Xia†, Lei Sha†, Baobao Chang† and Zhifang Sui†⋆ †Key Laboratory of Computational Linguistics (Ministry of Education), School of EECS, Peking University, 100871, Beijing, China ⋆Beijing Advanced Innovation Center for Imaging Technology, Capital Normal University, Beijing, China {xql,shalei,chbb,szf}@pku.edu.cn Abstract Previous studies on Chinese semantic role labeling (SRL) have concentrated on a sin- gle semantically annotated corpus. But the training data of single corpus is of- ten limited. Whereas the other existing semantically annotated corpora for Chi- nese SRL are scattered across different annotation frameworks. But still, Data sparsity remains a bottleneck. This sit- uation calls for larger training datasets, or effective approaches which can take advantage of highly heterogeneous data. In this paper, we focus mainly on the latter, that is, to improve Chinese SRL by using heterogeneous corpora together. We propose a novel progressive learn- ing model which augments the Progres- sive Neural Network with Gated Recur- rent Adapters. The model can accommo- date heterogeneous inputs and effectively transfer knowledge between them. We also release a new corpus, Chinese Sem- Bank, for Chinese SRL1. Experiments on CPB 1.0 show that our model outperforms state-of-the-art methods. 1 Introduction Semantic role labeling (SRL) is one of the fun- damental tasks in natural language processing be- cause of its important role in information extrac- tion (Bastianelli et al., 2013), statistical machine translation (Aziz et al., 2016; Xiong et al., 2012), and so on. However, state-of-the-art performance of Chi- nese SRL is still far from satisfactory. And data sparsity has been a bottleneck which can not be 1http://www.klcl.pku.edu.cn/ShowNews.aspx?id=156 Predicate given: 修改 revise (a) [ ArgM-TMP在这期间] Meanwhile , [Arg0全国人大常委会] the NPC Standing Committee ... 广泛 widely 征求 solicit 意见, opinions, [ArgM-ADV多次 for many times ] [ArgM-ADV反复 repeatedly ] [Rel修改 revise ] [Arg1*pro* (omitted) . ] 。 (b) [agent他们] They 对 to [patient系统 system ]进行了 made [Rel修改 revise . ] 。 Figure 1: Sentences from (a) CPB and (b) our het- erogeneous dataset. In CPB, each predicate (e.g., 修改) has a specific set of core roles given with numbers (e.g., Arg0). While our dataset uses a different semantic role set, and all roles are non- predicate-specific. ignored. For English, the most commonly used benchmark dataset PropBank (Xue and Palmer, 2003) has about 54,900 sentences. But for Chi- nese, there are only 10,364 sentences in Chinese PropBank 1.0 (CPB) (with about 35,700 proposi- tions) (Xue, 2008). To mitigate the data sparsity, models incor- porating heterogeneous resources have been in- troduced to improve Chinese SRL performance (Wang et al., 2015; Guo et al., 2016; Li et al., 2016). The heterogeneous resources introduced by these models include other semantically an- notated corpora …

Content truncated to 500 words